Draft Genetic Test Review

“This information is distributed solely for the purpose of pre dissemination peer review under applicable information quality guidelines. It has not been formally disseminated by the Centers for Disease Control and Prevention. It does not represent and should not be construed to represent any agency determination or policy.”

---

Breast Cancer
Disorder Setting

(98KB)

---

DISORDER/SETTING

Question 1: What is the specific clinical disorder to be studied?
Question 2: What are the clinical findings defining this disorder?
Question 3: What is the clinical setting in which the test is to be performed?
Question 4: What DNA test(s) are associated with this disorder?
Question 5: Are preliminary screening questions employed?
Question 6: Is it a stand-alone test or is it one of a series of tests?
Question 7: If it is part of a series of screening tests, are all tests performed in all instances (parallel) or are only some tests performed on the basis of other results (series)?

---

DISORDER/SETTING

Rationale for preliminary screening questions
Although breast cancer is relatively common, only a small proportion of such cases (Question 18) is associated with mutations detectable by direct sequencing of the BRCA1/2 genes. This factor, combined with the high cost of testing, provides the rationale for preliminary screening questions to identify appropriate candidates for genetic predisposition testing (Question 3). The aim of testing for BRCA1/2 mutations is to prevent the morbidity/mortality associated with breast (or ovarian) cancer by providing information to a population of high-risk individuals, so that informed decisions can be made regarding specific risk-reducing activities (Question 29). The areas queried include personal history of breast and/or ovarian cancer, age at diagnosis, family history of breast and/or ovarian cancer and age(s) at diagnosis, menopausal status, and whether the individual to be tested is Ashkenazi Jewish (Question 3). BRCA1/2 sequencing is not performed on individuals under 18 years of age except in unusual circumstances, as described by the American Society of Clinical Oncology (ASCO). A statement adopted by ASCO in 1996 recommended that breast/ovarian cancer predisposition testing be offered only in the setting of a "strong family history of cancer or very early age of onset of disease", further defined as at least a 10 percent probability of having a BRCA1/2 mutation. (1996) This threshold, though based on expert opinion, is arbitrary and subject to professional interpretation.  

A caveat of BRCA1/2 mutation testing is that variants of unknown clinical significance are identified in approximately 13 percent of all samples undergoing full sequencing. (Frank et al., 2002) Assuming that these variants are found in the same proportion of the general population, the number of these indeterminate test results would greatly surpass the number of deleterious mutations, if screening questions were not utilized. 

Models used to predict risk for carrying a BRCA1/2 mutation
BRCA1/2 are autosomal dominant genes, meaning that mutations can be inherited equally from the mother's or father's side of the family. Thus, family history and personal disease history increase the probability of finding a BRCA1/2 mutation in a woman. A possible hereditary risk of breast/ovarian cancer should be considered, if a family includes two or more women with breast cancer at an early age of onset (usually before age 50) and/or ovarian cancer at any age. (Armstrong et al., 2000; Frank et al., 1998) Race/ethnicity is also a consideration (i.e., the mutation prevalence is known to be increased among Ashkenazi Jewish woman). An older age at diagnosis is associated with a lower risk of finding a BRCA1/2 mutation.

Models have been developed to determine an individual’s a priori risk of carrying a BRCA1/2 mutation or to assess risk of breast cancer. Two models were developed to predict the probability of a BRCA1 mutation, though neither has been validated. (Berry et al., 1997; Couch et al., 1997) An extended model has subsequently been developed to predict the probability of both BRCA1 and BRCA2 mutations. (Parmigiani et al., 1998) This model has been developed into a computer program (BRCAPRO). BRCAPRO incorporates the autosomal dominant Mendelian characteristics of the genes, published prevalence and penetrance of BRCA1/2 mutations, and Bayesian methods. (Iversen et al., 2000) This program has been validated in a population at high risk for breast and/or ovarian cancer. (Berry et al., 2002; Euhus et al., 2002) Empiric data from BRCA1/2 mutation testing at Myriad Genetic Laboratories have been used to model the probability that an individual carries a BRCA1/2 mutation. (Frank et al., 1998; Shattuck-Eidens et al., 1997) Empiric models for predicting breast cancer risk have also been developed. (Claus et al., 1994; Gail et al., 1989; Houlston et al., 1992) Each of the above-listed models has strengths and weaknesses and is appropriate for use in certain settings. These models are reviewed in a recent publication. (Domchek et al., 2003) In addition, other methods are utilized in the clinical setting to assess risk of breast cancer and/or risk of carrying a BRCA1/2 mutation, including check lists provided by insurers or Myriad Genetic Laboratories. (Mackay, 1997) Women may be placed in different risk categories, depending on the method used to estimate risk. (Domchek et al., 2003; Tischkowitz et al., 2000) Given the current status of these models, it is important to involve an experienced health professional (e.g., a genetic counselor) to interpret risk estimates and provide counseling regarding BRCA1/2 mutation testing.

An example of data upon which these models are based is depicted in Table 1-2. The odds ratios of carrying a deleterious BRCA1 mutation are derived from a logistic regression model. (Shattuck-Eidens et al., 1997) According to Table 1-2, each year added to the age at diagnosis decreases the risk by 8%. As evidence of this effect, among a population-based sample of women under 35 years of age with breast cancer, unselected for family history, 6 of 80 (7.5%) had BRCA1 mutations. (Langston et al., 1996) Similar results were seen in another study, where 13 percent of women with very early onset breast cancer, and without a strong family history, had BRCA1 mutations. (FitzGerald et al., 1996) Both of these findings are higher than the expected 4 to 5% of BRCA1 mutations among women with breast cancer under age 55 in a general population. (Question 18).

Example of computing the risk of carrying a BRCA1 deleterious mutation

“The log odds (L) of an individual carrying a deleterious mutation is estimated by the following equation: L = -0.08a + 1.41b + 0.0c + 1.29d + 2.08e + 3.39f + 1.68g + 0.31h + 1.06i + 1.68j, where a is the age at diagnosis of breast and/or ovarian cancer; b is 1 if a patient is of Ashkenazi descent, 0 otherwise; c is 1 if the patient is diagnosed with unilateral breast cancer but not ovarian cancer, 0 otherwise (coefficient of c in the equation is 0 since this case is used as baseline, and it is included for completeness); d is 1 if the patient is diagnosed with bilateral breast cancer but not ovarian cancer, 0 otherwise; e is 1 if the patient is diagnosed with unilateral breast cancer and with ovarian cancer, 0 otherwise; f is 1 if the patient is diagnosed with bilateral breast cancer and with ovarian cancer, 0 otherwise; g is 1 if the patient is diagnosed with ovarian cancer but not breast cancer, 0 otherwise; h is number of relatives with breast cancer, but not ovarian cancer; i is number of relatives with ovarian cancer, but not breast cancer; and j is number of relatives with breast and ovarian cancer. The intercept was estimated to be 0.” (Shattuck-Eidens et al., 1997) The probability that an individual carries a BRCA1 mutation is: p = exp(L)/[1 + exp(L)]

Woman with a personal history of cancer Using the model described above, a 50 year old woman diagnosed with ovarian cancer and who has one relative with breast cancer is computed to have an 11.8 percent probability of having a deleterious BRCA1 mutation. (-2.01 = -0.08[50] + 1.68[1] + 0.31[1] and 0.118 = exp[-2.01]/[1 + exp(L)])

Woman without a personal history of cancer A woman with no personal history of breast or ovarian cancer who has 3 relatives with breast cancer and 1 relative with ovarian cancer is computed to have an 88 percent probability of having a deleterious BRCA1 mutation. (1.99 = 0.31[3] + 1.06[1] and 0.88 = exp[1.99]/[1 + exp(L)])

From (Shattuck-Eidens et al., 1997)

* Each year added to the age at diagnosis decreases the risk by 8%

Gap in Knowledge: Validation for specific models predicting BRCA1/2 risk. Although some studies have compared the risks predicted by different models, no study has compared the predicted risk for specific selected family histories versus the observed proportion of positive mutation studies found by Myriad Genetic Laboratories.

Accuracy of family history information – breast cancer
Accuracy of family history information for breast cancer has been investigated and is summarized in Table 1-3. Four of the six studies included only breast cancer patients or women who had been referred to a cancer genetics clinic. Accuracy of family history of breast cancer in the general population was assessed in the remaining two studies through the use of controls. These data are of limited use because sensitivity and specificity were not assessed in one study, and personal interview data were compared with those in a population database in the remaining study. This methodology is likely to underestimate sensitivity (the individual does indeed have cancer, but is not included in the registry). It would also likely result in the specificity being overestimated (some individuals not reporting cancer and not in the registry, do indeed have cancer, but were not included in the registry). Incorrect matching could result in over- or under-estimation of sensitivity and specificity. A single study estimated sensitivity and specificity by verifying reported cases of breast cancer with either pathology reports/clinical records, self-reports from the affected and non-affected relatives of the proband, or death certificates. Sensitivity refers to the proportion of reported cases of breast cancer among all cases. Sensitivity reported in two studies ranges from 83 to 95 percent. Specificity refers to the proportion of women reported not to have breast cancer among all those who do not have breast cancer. Specificity reported in three studies ranges from 93 to 99 percent. Positive predictive value is the proportion of women confirmed to have breast cancer among all those reported to have breast cancer. The positive predictive values ranged from 83 to 99 percent. Negative predictive value is the proportion of women without breast cancer among all those reported to not have breast cancer. This was assessed by studies 4 through 6 only. These studies reported a negative predictive value of approximately 98 percent. Figure 1-1 shows the impact of using a family history questionnaire in the screening process for identifying women at increased risk for carrying BRCA1/2 mutations. The following caveat should be considered. These estimates are based on the total number of reported cases, not on the number of individuals reporting cases. For example, if 35 women each correctly reported one first-degree relative with breast cancer but collectively failed to report two other cases, the sensitivity would be 95 percent (35/37). If these same 35 women each correctly reported two first-degree relatives with breast cancer but collectively failed to report 10 cases, then the sensitivity would be 88 percent (70/80).

N/A = Not Available

Reference: 1 (Love et al., 1985) , 2 (Parent et al., 1995) , 3 (Theis et al., 1994) , 4 (Ziogas and Anton-Culver, 2003) , 5 (Anton-Culver et al., 1996) , 6 (Kerber and Slattery, 1997)

Study 1. Wisconsin: Love et al. One hundred and twenty-one self-referred patients visiting a cancer prevention clinic at the University of Wisconsin provided a detailed history of cancers occurring in first-, second-, and third-degree relatives. Verification of a positive cancer family history was done by reviewing pathology and operative reports, hospital admission and discharge summaries, death certificates, and autopsy reports. Verification of negative cancer family history was not performed, thus sensitivity and specificity could not be calculated. Participants were correct in 91 percent (143/157, 95% CI 85.5-95.0%) of the cases for all relatives in whom they reported breast as the primary site, 94 percent (78/83, 95% CI 86.5-98.0%) of the cases in first-degree relatives, and 88 percent (65/74, 95% CI 78.2-94.3%) of the cases in second- and third-degree relatives.

Study 2. Canada: Parent et al. reported 414 French-Canadian women recently diagnosed with primary breast cancer and 429 age-matched population-based controls, all of whom provided information on relatives affected with any type of cancer. A total of 105 women (68 cases and 37 controls) reported a history of breast cancer in at least one first-degree relative. The accuracy was confirmed via pathological records. Cases correctly reported 74 out of 81 first-degree relatives with breast cancer (positive predictive value of 89 percent - 95% CI 83.0-96.4%), while controls were correct in 33 out of 34 (positive predictive value of 97 percent - 95% CI 84.7-99.9%). The overall positive predictive value was 93 percent (95% CI 86.8-97.0%). Sensitivity and specificity were not assessed. Overall, 11 percent of reports contained errors of more than five years from the real age at diagnosis.

Study 3. Canada: Theis et al. reported on 165 breast cancer patients in a Toronto hospital who provided family cancer histories in first- and second-degree relatives. Of the 186 reported cases of breast cancer in first-degree relatives, 167 records were obtained. Confirmation of this diagnosis was made in 166 cases (positive predictive value of 99.4 percent - 95% CI 96.7-99.99). In second-degree relatives, 33 of 39 reported breast cancer cases were correctly identified (positive predictive value of 84.6% - 95% CI 69.5-94.1). Specificity was assessed by randomly sampling 100 first-degree relatives reported as cancer-free. None of these relatives appeared in the Ontario cancer registry and were assumed to not have cancer (specificity = 99 percent, 95% CI 94.6-99.98%). Data for ovarian cancer were sparse. Only two cases were reported and had records obtained. Both cases were confirmed.

Study 4. California: Ziogas et al. studied 670 cases of breast cancer in Orange County, California. Of these cases, 638 were population-based and 32 were clinic-based. Eight male breast cancer cases are included. Validation of family history of breast cancer was done by comparing data obtained by personal interview with pathology reports (474), self-reports (777), or death certificates (2142) on the relatives. The sensitivity of the case individuals’ report of their first-degree relatives’ histories of breast cancer was 95.4 percent (95 percent CI 92.6-98.3%). The specificity was 97.4 percent (95 percent CI 96.4-98.4). Of the 211 cases of breast cancer reported in the interviews, 188 were confirmed by one of the reference standards (positive predictive value of 89.1 percent (95 percent CI 84.1-93.0%). Predictors of false negative reports of breast cancer were age greater than 70 years, and reports of cancer in 2^nd and 3^rd degree relatives. Predictors of false positive reports were not broken down by proband cancer type. For all cancers combined, false positives were more likely to be reported by males and clinic-based probands

Study 5. California: Anton-Culver et al. validated family history of breast cancer reported by 359 breast cancer probands in Orange County with data contained in a cancer registry. This cancer registry is one of the ten in the California Cancer Reporting System and meets all reporting requirements of the Surveillance, Epidemiology, and End Results program of the National Cancer Institute. Ascertainment of cases has been shown to be 97 percent complete. Using the cancer registry as the standard, the sensitivity of the personal interview data on breast cancer history in mothers and sisters was 91.4% (95% CI 81.0-97.1). The specificity was 98.4% (95% CI 96.5-99.4). Of the 59 cases of breast cancer reported in the interview, 53 were confirmed by the registry (PPV=89.8%, 95% CI 79.2-96.5).

Study 6. Utah: Kerber and Slattery reported on 881 cases and controls from the Diet, Activity, and Reproduction in Colon Cancer study. (Kerber and Slattery, 1997) Of these, 331 (37.6%) could be linked to the Utah Population Database (UPDB), which contains genealogic and cancer information. The proportion of the Utah population in the UPDB falls from about 60 percent between 1920 and 1934 to just over 30 percent by 1960. A comparison was made between self-reporting of family history of breast cancer and data in the UPDB. Sensitivity and specificity for first-degree relative reporting of breast cancer were 82.9 percent (95% CI 66.4-93.4%) and 92.6 percent (95% CI 89.0-95.3%), respectively. Sensitivity and specificity were slightly higher in cases (84.6 and 95.5%, respectively) than in controls (81.8 and 90.8%, respectively). Of the 51 cases of breast cancer reported by participants, 29 were confirmed by the UPDB (positive predictive value of 56.9 percent, 95% CI 42.2-70.6). The positive predictive value for reporting breast cancer cases was 68.7 percent in cases and 51.4 percent in controls.

Studies not Included: Another study utilized family history information from 408 confirmed family cancer case notes in two regional cancer genetics departments. Information from cancer registries, death certificates, hospital notes, and histopathological records were used to confirm reported family history of breast cancer. (Douglas et al., 1999) The accuracy of breast cancer family history was 94 percent. Verification of negative history was not reported. Because no raw numbers or other data were given, this study could not be combined with those in Table 1-2.

Two studies have reported the validation of a personal history of cancers. In the first, the validity of self-reported breast cancer diagnosis (personal history) was compared with population-based cancer registry data in 65,582 men and women aged 39 to 96 years, who were participants in the Cancer Prevention Study II Nutrition survey. (Bergmann et al., 1998) Sensitivity was 91 percent (779/853, 95 percent CI 89.2-93.1%) and specificity was 99.8 percent (64,587/64,729, 95 percent CI 99.7-99.8%) in breast cancer personal history reporting. Positive predictive value was 84.6 percent (95 percent CI 82.1-86.9%). The second study validated self-reported cancers from the California Teachers Study with the California Cancer Registry. (Parikh-Patel et al., 2003) Of the 121,196 teachers included in the study, 3,103 were found in the registry to have breast cancer. Only 2,991 of these teachers reported a personal history of breast cancer (sensitivity = 96.4%, 95% CI 95.6-97.5). Among the 118,093 teachers who did not have a breast cancer found in the registry, 115,849 reported a negative personal history (specificity = 98.1%, 95% CI 98.1-98.2); the remaining 2,244 falsely reported a positive personal history of breast cancer. The positive predictive value was 57.1 percent (95 percent CI 55.8-58.5) and negative predictive value was 99.9 percent. The only statistically significant predictor of accurate reporting was age of less than 45 years. An additional statistically significant predictor of false negative reports was in situ stage of cancer at diagnosis (OR = 8.22, 95 percent CI 5.4-12.5).

Gap in Knowledge: Reliability of Sensitivity and Specificity of Family History Questionnaires. Data provided in Table 1-3 show heterogeneity in estimates of sensitivity and specificity. Data from studies 4 and 5 are based on the assumption that cancer registries are 100% accurate. This is unlikely to be true. Incomplete ascertainment will likely cause sensitivity to be underestimated (the individual does indeed have cancer, but is not included in the registry). It would also likely result in the specificity being overestimated (some individuals not reporting cancer and not in the registry, do indeed have cancer, but were not included in the registry). Incorrect matching could result in over- or underestimation of sensitivity and specificity.

Assumptions: Prevalence of family history is 6.2% (Question 19, Appendix A)
                    Sensitivity of family history questionnaire is 91%.

Accuracy of family history information – Ovarian cancer
Limited data are available regarding the validation of ovarian cancer family history. Validation of family history of ovarian cancer was done by comparing data obtained from personal interview with pathology reports, self-reports, or death certificates on the relatives. (Ziogas and Anton-Culver, 2003) Sensitivity and specificity for first-degree relative reporting of ovarian cancer were 83.3 percent (95 percent CI, 68.6-93.0%) and 98.9 percent (95 percent CI, 98.1-99.5%), respectively. The positive predictive value was 76.1 percent (95 percent CI, 61.2-87.4%). Self-reporting of family history of ovarian cancer was compared to genealogic and cancer information in the Utah Population Database. (Kerber and Slattery, 1997) Sensitivity and specificity for first-degree relative reporting of ovarian cancer were 60 percent (95 percent CI, 14.7-94.7%) and 97.6 percent (95 percent CI, 95.2-98.9%), respectively. The positive predictive value was 27.3 percent (95 percent CI, 6.0-61.0%). A study in the UK utilized information from cancer registries, death certificates, hospital notes, and histopathological records to confirm reported family history of ovarian cancer. (Douglas et al., 1999) The positive predictive value of ovarian cancer family history was 83 percent. Verification of negative history was not reported.